Scaling issues in mapping riparian zones with remote sensing data: Quantifying errors and sources of uncertainty

Riparian zones are the ecotones or transition areas between upland and aquatic ecosystems, located at the margins of rivers, lakes, ponds and wetlands. Their boundaries are defined by changes in soil, moisture, and vegetation (Naiman 2000, D?camps 1996, Gregory et al. 1991). Although these ecosystems may be small relative to the aquatic systems they abut, they perform many important ecosystem services, including shading (thus buffering air and water temperature), retaining nutrients and/or sediments, stabilizing stream banks and littoral zones, and providing organic material (leaves, wood) and critical habitat for a diverse community of plant and animal species (Malanson 1993). Riparian zones are highly variable systems whose structure and composition are shaped by geomorphology, vegetation patterns, disturbance regimes (D?camps 1996), as well as current (Erickson and DeYoung 1993) and perhaps historic land use practices (Foster et al. 2003). Processes that operate over a large range of temporal and spatial scales control these structuring factors. At one end of the time/space continuum are processes such as tectonics, volcanism, glaciation, and climate change. At intermediate spatial and temporal scales are historic land use practices (e.g., burning regimes implemented by native peoples, permanent land cover conversion) and catastrophic flooding. At small scales, localized flooding and land management practices influence the structure and function of riparian zones. Some processes occur over multiple scales and their effects may also vary by scale. The fine-scale variation of vegetative cover resulting from moisture and soil gradients around streams and wetlands has posed a challenge to research scientists and land managers (Muller 1997, Congalton et al. 2002). One of the current challenges in watershed science is the development of economical and accurate methods for mapping riparian areas. The problem of quantifying riparian zones is particularly difficult in developed areas (e.g., urban and agricultural) and arid landscapes, where riparian zones are frequently narrower than the resolution of the standard land cover data sets often used for watershed characterization. The current widespread availability of inexpensive and user-friendly geographic information systems (GIS), and a larger selection of spatial data layers with global to regional coverage has expanded the use of these tools worldwide. A negative effect of the expanded use of these tools and data is that users occasionally overlook the inherent limitations, particularly with respect to the scale (grain and extent) of the data relative to the question being asked. The recent availability of an updated USGS National Land Cover Dataset (NLCD; Vogelmann et al. 2001) has enabled researchers, managers and policy makers to explore many new questions. Of particular interest is the possible use of the NLCD for addressing water quality questions from a watershed perspective. The specific requirements for mapping vegetation communities from satellite imagery were assessed by Woodcock and Strahler (1987) and Marceau (1994). Both papers concluded that satellite remote sensing was more appropriate for mapping broad vegetative categories rather than distinct vegetative communities. Although higher resolution imagery is now available, the high cost of acquiring and processing these data precludes their use for large regions. Yet, high-resolution data derived from aerial photography also is constrained by the high cost of acquiring, interpreting, and digitizing data. Therefore, highly resolved maps of riparian areas are often only available for portions of the streams, as opposed to the entire contributing watershed. Muller (1997, p. 419) states, "The scale factor is of prior importance for studying riparian vegetation," yet this factor is seldom taken into account in ecological assessments of riparian zone function. Our objectives in this chapter are to (1) quantify the potential errors and sources of uncertainty in the use of the NLCD data for mapping land use and cover in riparian zones in agricultural landscapes, (2) examine the scale effects with respect to changing grain and extent in these mapping exercises, (3) compare these phenomena in stream versus wetland riparian ecosystems, and (4) assess these differences in landscapes whose matrix is dominated by agriculture versus those with a more diverse matrix. Quantifying scale effects, hierarchical linkages, and resulting uncertainty associated with mapping and analyzing riparian zones is necessary for understanding the structure and composition of riparian areas along streams and wetlands. An understanding of the underlying processes that influence riparian zones and control water quality is of fundamental interest not only to researchers and resource managers, but also to policy makers and private landowners who must evaluate the potential consequences of land management practices under different scenarios.